| Author | Message | ||
     Nuno Matos (Nmatos) |
Try to use the Mahalanobis distance on the space defined by the PCA scores. Your calibration set will be composed only by "good" samples. The external validation set would be "good" and "bad" samples. Using this same model you can turn it into a biparametric discrimination model in the way that you can reconctruct your spectra using PCA and, therefore, obtaining spectral residuals. Hope this help | ||
| Lazzar One |
Dear all, I have to prepare a calibration to disriminate between "good MEG" ( monoethylene glycol ) and "bad MEG " . The problem is that there are more many way not to be the right quality MEG than there are to be right quality MEG. Should I use a two level Y variable where good MEG sample will be assigned a value of 1 and all the rest of the sample a value of 0 or should I try to discriminate more within the bad samples and add more level to my variable or should I use more variables ? Thanks |